The Kaplan-Meier estimator of a survival function is well-known to be asymptotically efficient when cause of failure (censored or non-censored) is always observed. We consider the problem of finding an estimator when the failure indicators are missing completely at random. Under this assumption,...